Publication Date

2012-07-13

Availability

Open access

Embargo Period

2012-07-13

Degree Type

Thesis

Degree Name

Master of Science (MS)

Department

Electrical and Computer Engineering (Engineering)

Date of Defense

2012-06-20

First Committee Member

Mei-Ling Shyu

Second Committee Member

Saman Aliari Zonouz

Third Committee Member

Frank D. Marks, Jr.

Abstract

Accurate forecasting of Tropical Cyclone (TC) track and intensity are vital for safeguarding the lives and property of communities in regions that are subject to TC impact. While there have been impressive advancements in TC track forecasting over the last 40 years, forecasts of TC intensity have seen virtually no improvement since 1990, chiefly because of the difficulty of predicting rapid changes in TC intensity. This study applies data mining techniques to a data set of meteorological parameters in order to construct an associative classifier that has been named AprioriGrad. This classifier is based on the association rule mining technique together with the Apriori algorithm for frequent itemset selection, but includes customizations for detecting rare events and for labeling a series of interrelated classification targets defined as yes/no thresholds on an underlying continuous measurement of 24-h TC intensity change. AprioriGrad’s performance on this domain is compared to a variety of classification techniques, and implications for possible development as an operational forecasting tool or for further meteorological study are examined.

Keywords

data mining; associative classification; rare events prediction; Apriori algorithm; tropical cyclone intensity; rapid intensification

Download

Share

COinS